Naphula/Delirium-v1-abliterated
Abliterated vs sam-paech/Delirium-v1. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 26 of 42 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/1, Norms 0/169, MLP 26/126, Attention 26/168
2 derivatives checked
- mradermacher/Delirium-v1-abliterated-i1-GGUF: Quantization
- mradermacher/Delirium-v1-abliterated-GGUF: Same weights, converted